Download the full version of the ebook at ebookname.

com

Quaternion Fourier Transforms for Signal and Image

Processing 1st Edition Todd A. Ell

https://ebookname.com/product/quaternion-fourier-transforms-
for-signal-and-image-processing-1st-edition-todd-a-ell/

OR CLICK BUTTON

DOWNLOAD EBOOK

Download more ebook instantly today at https://ebookname.com

Instant digital products (PDF, ePub, MOBI) available
Download now and explore formats that suit you...

Digital Filters Design for Signal and Image Processing

Mohamed Najim

https://ebookname.com/product/digital-filters-design-for-signal-and-
image-processing-mohamed-najim/

ebookname.com

Signal and Image Processing for Remote Sensing 1st Edition

C.H. Chen

https://ebookname.com/product/signal-and-image-processing-for-remote-
sensing-1st-edition-c-h-chen/

ebookname.com

Signal and Image Processing in Navigational Systems

Electrical Engineering Applied Signal Processing Series
1st Edition Vyacheslav Tuzlukov
https://ebookname.com/product/signal-and-image-processing-in-
navigational-systems-electrical-engineering-applied-signal-processing-
series-1st-edition-vyacheslav-tuzlukov/
ebookname.com

■■■■■■ In Love with Fruit Desserts Chinese English Edition

■■■

https://ebookname.com/product/%e6%84%9b%e4%b8%8a%e6%b0%b4%e6%9e%9c%e7%
94%9c%e9%bb%9e-in-love-with-fruit-desserts-chinese-english-
edition-%e4%bd%95%e5%b9%bc%e6%97%8b/
ebookname.com
Return Distributions in Finance Quantitative Finance
Quantitative Finance First Edition Stephen Satchell

https://ebookname.com/product/return-distributions-in-finance-
quantitative-finance-quantitative-finance-first-edition-stephen-
satchell/
ebookname.com

Suzy Gershman s Born to Shop Hong Kong Shanghai Beijing

The Ultimate Guide for People Who Love to Shop Born To
Shop 4th edition Suzy Gershman
https://ebookname.com/product/suzy-gershman-s-born-to-shop-hong-kong-
shanghai-beijing-the-ultimate-guide-for-people-who-love-to-shop-born-
to-shop-4th-edition-suzy-gershman/
ebookname.com

Grace Full Leadership Understanding the Heart of a

Christian Leader 1st Edition John Bowling

https://ebookname.com/product/grace-full-leadership-understanding-the-
heart-of-a-christian-leader-1st-edition-john-bowling/

ebookname.com

MCSA MCSE Self Paced Training Kit Exam 70 291 Implementing

Managing and Maintaining a Microsoft Windows Server 2003
Network Infrastructure Pro Certification 2nd ed. Edition
J.C. Mackin
https://ebookname.com/product/mcsa-mcse-self-paced-training-kit-
exam-70-291-implementing-managing-and-maintaining-a-microsoft-windows-
server-2003-network-infrastructure-pro-certification-2nd-ed-edition-j-
c-mackin/
ebookname.com

The Pfeiffer Book of Successful Team Building Tools Best

of the Annuals 2nd Edition Elaine Biech

https://ebookname.com/product/the-pfeiffer-book-of-successful-team-
building-tools-best-of-the-annuals-2nd-edition-elaine-biech/

ebookname.com
Weighting Evidence in Language and Literature A
Statistical Approach 1st Edition Barron Brainerd

https://ebookname.com/product/weighting-evidence-in-language-and-
literature-a-statistical-approach-1st-edition-barron-brainerd/

ebookname.com
W478-Sangwine.qxp_Layout 1 10/04/2014 16:25 Page 1

FOCUS SERIES in DIGITAL SIGNAL AND IMAGE PROCESSING

FOCUS

Stephen J. Sangwine
Nicolas Le Bihan
Todd A. Ell
DIGITAL SIGNAL AND IMAGE PROCESSING SERIES
This book presents the state of the art, together with the most recent
research results, in the use of Quaternion Fourier Transforms (QFT) for
the processing of color images and complex valued signals. It is based
on the work of the authors in this area since the 1990s and presents the
mathematical concepts, computational issues and applications on
images and signals. The book, together with the MATLAB toolbox
developed by two of the authors (QTFM, http://qtfm.sourceforge.net/),

Quaternion Fourier Transforms for Signal and Image Processing

allows the reader to make use of the presented concepts and
experiment with them in practice through the examples provided in the
book.
Following the Introduction, Chapter 1 introduces the quaternion algebra
H and presents some properties which will be of use in the subsequent
chapters. Chapter 2 gives an overview of the geometric transformations
which can be represented using quaternions. Chapter 3 provides the
definition and properties of QFT. The signals and images considered are
those with vector-valued samples/pixels.
The fourth and final chapter is dedicated to the illustration of the use of
QFT to process color images and complex improper signals. The
concepts presented in this chapter are illustrated on simulated and real
images and signals.

Todd A. Ell is an Engineering Fellow at UTC Aerospace Systems,

Burnsville, MN, USA. He is also a Visiting Fellow at the University of
Quaternion Fourier
Essex, Colchester, UK. His interests include the study and application of
hypercomplex algebras to dynamic systems analysis.
Nicolas Le Bihan is currently a Chargé de Recherche at the CNRS
Transforms for Signal
working in the Department of Images & Signals (DIS) at the GIPSA-Lab
in Grenoble, France. His research interests include polarized signal
processing, statistical signal processing on groups and manifolds,
and Image Processing
geometric (Berry) phases, random processes on non-commutative
algebraic structures and applications in physics and geophysics.
Stephen J. Sangwine is a Senior Lecturer with the School of Computer
Science and Electronic Engineering, University of Essex, Colchester, UK.
His interests include linear vector filtering and transforms of vector Todd A. Ell, Nicolas Le Bihan
signals and images, color image processing, and digital hardware
design. and Stephen J. Sangwine

www.iste.co.uk
Z(7ib8e8-CBEHIB(
Quaternion Fourier Transforms for
Signal and Image Processing
 FOCUS SERIES

Series Editor Francis Castanié

Quaternion Fourier
Transforms for Signal
and Image Processing

Todd A. Ell
Nicolas Le Bihan
Stephen J. Sangwine
First published 2014 in Great Britain and the United States by ISTE Ltd and John Wiley & Sons, Inc.

Apart from any fair dealing for the purposes of research or private study, or criticism or review, as
permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced,
stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers,
or in the case of reprographic reproduction in accordance with the terms and licenses issued by the
CLA. Enquiries concerning reproduction outside these terms should be sent to the publishers at the
undermentioned address:

ISTE Ltd John Wiley & Sons, Inc.

27-37 St George’s Road 111 River Street
London SW19 4EU Hoboken, NJ 07030
UK USA

www.iste.co.uk www.wiley.com

© ISTE Ltd 2014

The rights of Todd A. Ell, Nicolas Le Bihan and Stephen J. Sangwine to be identified as the author of this
work have been asserted by them in accordance with the Copyright, Designs and Patents Act 1988.

Library of Congress Control Number: 2014934161

British Library Cataloguing-in-Publication Data

A CIP record for this book is available from the British Library
ISSN 2051-2481 (Print)
ISSN 2051-249X (Online)
ISBN 978-1-84821-478-1

Printed and bound in Great Britain by CPI Group (UK) Ltd., Croydon, Surrey CR0 4YY
 Contents

N OMENCLATURE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

P REFACE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

I NTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii

C HAPTER 1. Q UATERNION A LGEBRA . . . . . . . . . . . . . . . . . . . . . 1

1.1. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2. Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3. Exponential and logarithm of a quaternion . . . . . . . . . . . . . . . . 7
1.3.1. Exponential of a pure quaternion . . . . . . . . . . . . . . . . . . . . 7
1.3.2. Exponential of a full quaternion . . . . . . . . . . . . . . . . . . . . 9
1.3.3. Logarithm of a quaternion . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4. Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.1. Polar forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.2. The Cj -pair notation . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4.3. R and C matrix representations . . . . . . . . . . . . . . . . . . . . . 17
1.5. Powers of a quaternion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.6. Subfields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

C HAPTER 2. G EOMETRIC A PPLICATIONS . . . . . . . . . . . . . . . . . . . 21

2.1. Euclidean geometry (3D and 4D) . . . . . . . . . . . . . . . . . . . . . . 21
2.1.1. 3D reflections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.1.2. 3D rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.1.3. 3D shears . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.1.4. 3D dilations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
vi Quaternion Fourier Transforms for Signal and Image Processing

2.1.5. 4D reflections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.6. 4D rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2. Spherical geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3. Projective space (3D) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.1. Systems of linear quaternion functions . . . . . . . . . . . . . . . . 31
2.3.2. Projective transformations . . . . . . . . . . . . . . . . . . . . . . . . 33

C HAPTER 3. Q UATERNION F OURIER T RANSFORMS . . . . . . . . . . . . 35

3.1. 1D quaternion Fourier transforms . . . . . . . . . . . . . . . . . . . . . 38
3.1.1. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.1.2. Basic transform pairs . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.1.3. Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.1.4. Inter-relationships between definitions . . . . . . . . . . . . . . . . . 45
3.1.5. Convolution and correlation theorems . . . . . . . . . . . . . . . . . 47
3.2. 2D quaternion Fourier transforms . . . . . . . . . . . . . . . . . . . . . 48
3.2.1. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.2.2. Basic transform pairs . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2.3. Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.2.4. Inter-relationships between definitions . . . . . . . . . . . . . . . . . 55
3.3. Computational aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.1. Coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.2. Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.3.3. Verification of transforms . . . . . . . . . . . . . . . . . . . . . . . . 62

C HAPTER 4. S IGNAL AND I MAGE P ROCESSING . . . . . . . . . . . . . . . 67

4.1. Generalized convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.1.1. Classical grayscale image convolution filters . . . . . . . . . . . . . 67
4.1.2. Color images as quaternion arrays . . . . . . . . . . . . . . . . . . . 70
4.1.3. Quaternion convolution . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.1.4. Quaternion image spectrum . . . . . . . . . . . . . . . . . . . . . . . 73
4.2. Generalized correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2.1. Classical correlation and phase correlation . . . . . . . . . . . . . . 81
4.2.2. Quaternion correlation . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.2.3. Quaternion phase correlation . . . . . . . . . . . . . . . . . . . . . . 88
4.3. Instantaneous phase and amplitude of complex signals . . . . . . . . . 91
4.3.1. Important properties of 1D QFT of a complex signal z(t) . . . . . . 91
4.3.2. Hilbert transform and right-sided quaternion spectrum . . . . . . . 96
4.3.3. The quaternion signal associated with a complex signal . . . . . . . 98
4.3.4. Instantaneous amplitude and phase . . . . . . . . . . . . . . . . . . . 101
4.3.5. The instantaneous frequency of a complex signal . . . . . . . . . . . 102
 Contents vii

4.3.6. Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

4.3.7. The quaternion Wigner-Ville distribution of a complex signal . . . 109
4.3.8. Time marginal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.3.9. The mean frequency formula . . . . . . . . . . . . . . . . . . . . . . 113

B IBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

I NDEX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
 Nomenclature

Symbol Meaning Section Equation

R Set of real numbers section 1.1
C Set of complex numbers section 1.1
H Set of quaternions section 1.1
V(H) Set of pure quaternions section 1.1
Cμ Subfield of H section 1.6
I Complex root of −1 (See note below.)
(.) Real part section 1.1
(.) Imaginary part section 1.1
i, j, k Quaternion basis elements section 1.1 [1.2]
μ Pure quaternion section 1.1 [1.2]
i i-imaginary part section 1.1 [1.4]
j j-imaginary part section 1.1 [1.4]
k k-imaginary part section 1.1 [1.4]
MR (.) Real 4 × 4 matrix representation section 1.4 [1.69]
MC (.) Complex 2 × 2 matrix representation section 1.4 [1.70]
[[.]] Alternate real 4 × 4 matrix representation section 2.3 [2.20]
[.] Real 4 × 1 vector representation section 2.3 [2.20]
S(.) Scalar part section 1.1 [1.3]
V(.) Vector part section 1.1 [1.3]
(q1 , q2 ) Cj -pair notation of q section 1.4
AB Arc of great circle between A and B section 2.2
q Quaternion conjugate section 1.2 [1.18]
z Complex conjugate section 1.4 [1.54]
qμ Involution section 1.2 [1.23]
q −1 Inverse section 1.2 [1.26]
p, q Inner product section 1.2 [1.10]
μ×η Cross product section 1.2 [1.9]
μ⊥η Orthogonality section 1.2
x Quaternion Fourier Transforms for Signal and Image Processing

q p Bicomplex product section 1.4 [1.65]

{1, i, j, k}
Canonical basis in H section 1.1
q Norm section 1.2 [1.14]
|q| Modulus section 1.2 [1.16]
q̃ Unit quaternion |q̃| = 1 section 1.2 [1.30]
f ∗g Convolution section 3 [3.1]
f g Correlation section 3 [3.1]
F {.} Fourier transform section 3 [3.1]
Pη [.] Reflection operator section 2.1 [2.1]
Rq [.] Rotation operator section 2.1 [2.3]
Sα,β,μ [.]Shear operator section 2.1 [2.9]
Dμ,α [.] Dilation operator section 2.1 [2.10]
L1 (G; K) Space of absolutely integrable K-valued
functions taking arguments in G section 3.1
L2 (G; K) Space of square integrable K-valued
functions taking arguments in G section 4.3
sgn(.) Signum function section 4.3 [4.31]
p.v.(.) Principal value of an integral section 4.3 [4.32]

N OTE.– The complex root of −1 which is usually denoted i, or, in engineering texts
j, is denoted throughout this book by a capital letter I, in order to avoid any
confusion with the first of the three quaternion roots of −1, all three of which are
denoted throughout in bold font like this: i, j, k.
 Preface

This book aims to present the state of the art, together with the most recent
research results in the use of quaternion Fourier transforms (QFTs) for the processing
of color images and complex-valued signals. It is based on the work of the authors in
this area since the 1990s and presents the mathematical concepts, computational
issues and some applications to signals and images. The book, together with the
MATLAB® toolbox developed by the authors, [SAN 13b] allows the readers to
make use of the presented concepts and experiment with them in practice through the
examples provided.

Todd A. E LL
Minneapolis, MN, USA
Nicolas le B IHAN
Melbourne, Australia
Stephen J. S ANGWINE
Colchester, UK
April 2014
 Introduction

This book covers a topic that combines two branches of mathematical theory to
provide practical tools for the analysis and processing of signals (or images) with
three- or four-dimensional samples (or pixels). The two branches of mathematics are
not recent developments, but their combination has occurred only within the last
25–30 years, and mostly since just before the millennium.

I.1. Fourier analysis

Fourier analysis was, in 1822, with Joseph Fourier’s development of techniques,

the first to analyze mathematical functions into sinusoidal components. In signal and
image processing, Fourier’s ideas underpin the two fundamental representations of a
signal: one in the time (or image) domain where the signal (or image) is represented
by samples (or pixels) with amplitudes and the other in the frequency domain where
the signal (or image) is represented by sinusoidal frequency components, each with
an amplitude and a phase. Mathematically, these concepts are not limited to time and
frequency: one can use Fourier analysis on a function of any variable, resulting in a
representation in terms of sinusoidal functions of that variable. However, this book is
concerned with signal and image processing, and we will therefore use the terms time
and frequency rather than more general concepts. It should be understood throughout
that when we talk of images, the concept of time is replaced by the two spatial
coordinates that define pixel position within an image.

Today, Fourier analysis is classically taught to mathematicians, scientists and

engineers in several related ways, each applicable to a specific subset of
mathematical functions or signals:
1) Fourier series analysis [SNE 61] in which continuous periodic functions of
time, with infinite duration, are represented as sums of cosine and sine functions,
each with infinite duration;
xiv Quaternion Fourier Transforms for Signal and Image Processing

– Fourier integrals or transforms [BRA 00, ROB 68] in which continuous

(but aperiodic) functions of time are represented as continuous functions of frequency
(or vice versa);

2) Discrete Fourier transforms in which signals defined at discrete intervals in time

are represented in the frequency domain by cosine and sine functions. This topic is
broken down into:
– discrete-time Fourier transforms, in which discrete-time signals of limited
duration are represented as continuous frequency-domain distributions;
– discrete Fourier transforms, in which discrete-time, discretized (that is
digital) signals of finite duration are represented by a finite-length array of digital
frequency coefficients. (These are usually computed numerically using the fast Fourier
transform (FFT)).

The key to all of the above ideas is the representation of a signal using complex
exponentials, often known as harmonic analysis, although this term has a somewhat
wider meaning in mathematics than its usage in signal and image processing. The
complex exponential with angular frequency ω and phase φ: f (t) = A exp(ωt +
φ) = A (cos(ωt + φ) + I sin(ωt + φ)) has cosine and sine components in its real
and imaginary parts, respectively. Since, in this book, we are concerned with signals
that have three- or four-dimensional samples, it is helpful to consider classical Fourier
analysis in terms of complex exponentials rather than in terms of separate cosines and
sines.

Figure I.1 shows a real-valued signal (on the left-hand side of the plot, with time
increasing away from the viewer). The signal is a sawtooth waveform reconstructed
from its first five non-zero harmonics, which are plotted in the center of the figure
as helices. (The horizontal spacing between the helices is introduced simply to make
them clearer: there is no mathematical significance to it). The five helices on the left
are the positive frequency complex exponentials and the five helices on the right are
the negative frequencies. Note that the positive and negative frequency exponentials
have opposite directions of rotation. The real parts of the harmonics are projected onto
the right-hand side of the figure (these sum to give the reconstructed waveform on the
left) and the imaginary parts of the harmonics are projected onto the base of the figure
(these cancel out because the exponentials occur in complex conjugate pairs at positive
and negative frequencies, a symmetry due to the original signal being real-valued).

In general, with a complex signal analyzed into complex exponentials in the same
way, there would be no symmetry between the positive and negative frequency
exponentials. This case is a useful model for what follows in this book, where we
consider signals and images with three- and four-dimensional samples. Figure I.2
shows a complex signal constructed by bandlimiting a random complex signal.
 Introduction xv

Time is plotted on the right, increasing to the right, and at each time instant the
signal has a complex value. The signal evolution over time traces out a path in the
complex plane, and the figure renders this path as a three-dimensional view by
plotting the signal values, in effect, on a stack of 2,000 transparent complex planes
perpendicular to the time axis. The real and imaginary parts of the signal are also
plotted on the base of the axes, and on the rear plane of the axes. Analysis of a
complex signal into positive and negative frequency complex exponentials is not
conceptually different from the real case depicted in Figure I.1: each complex
exponential will have an amplitude and phase, and their sum will reconstruct the
original signal.

300

200

100
Amplitude

−100

−200

−300
2000

1500

1000

500

0
Time

Figure I.1. Analysis of a real signal into complex exponential harmonics

The time and frequency domain representations of a signal are not mutually
exclusive: the field of time-frequency analysis [FLA 98] is concerned with
intermediate representations that combine aspects of time and frequency. The need
for intermediate representations arises due to the variation of frequency content in a
signal over time. This is not an easy concept to understand, but it follows from the
uncertainty principle or Gabor limit: a signal cannot be bandlimited (i.e. with
frequency content limited to a finite range of frequencies) and simultaneously be of
limited time duration. A pure sinusoidal signal with unlimited duration (infinite
xvi Quaternion Fourier Transforms for Signal and Image Processing

extent) can be represented in the frequency domain as an impulse (that is a function

with zero value everywhere except at one frequency point). Conversely, an impulse in
the time domain has infinite bandwidth. However, a signal that contains a specific
frequency for a limited time requires a time-frequency representation. Examples of
such signals occur widely in the real world: speech and music contain frequencies
that are present for a short time (one note played on a musical instrument, for
example, which lasts for the duration of the note, plus some reverberation time
afterward). An in-depth discussion of these ideas is outside the scope of this book,
but is assumed to be understood; although much of the contents of the book relates to
Fourier transforms, the quaternion approach can easily be applied to time-frequency
concepts, such as fractional and short-time Fourier transforms, by combining
quaternion transform formulations with existing knowledge from classical signal
processing.

0.3

0.2

0.1

0
ℑ(f(t))

−0.1

−0.2

−0.3

−0.4
−0.2
2000
−0.1
1500
0
1000
0.1 500
0.2 0
t
ℜ(f(t))

Figure I.2. A bandlimited complex signal showing real and imaginary parts
projected onto the base and rear of the grid box

I.2. Quaternions

In this book, we are concerned with signals and images that have vector-valued
samples (that is samples with three or more components), and their processing using
 Introduction xvii

Fourier transforms based on four-dimensional hypercomplex numbers (quaternions).

In Chapter 4 (section 4.3), we show that quaternion Fourier transforms also have
applications for the processing of complex signals, exploiting the symmetry
properties of a quaternion Fourier transform that are missing from a complex Fourier
transform.

A vector-valued signal (in three dimensions, for example) evolves over time and
traces out a path in three-dimensional space. To render a plot of such a signal requires
four dimensions, and therefore we cannot produce a graphical representation like the
one in Figure I.2. Decomposition of a vector-valued signal into harmonic components
requires a Fourier transform in an algebra with dimension higher than 2, and this is
the motivation for the use of quaternions, which, as we will see, are the next available
higher-dimensional algebra after the complex numbers.

Quaternions followed the work of Fourier just over 20 years later, Sir William
Rowan Hamilton in 1843 to generalize the complex numbers to three dimensions,
was forced to resort to four dimensions in order to obtain what we now call a normed
division algebra, that is, an algebra where the norm of a product equals the product
of the norms, and where every element of the algebra (except zero) has a
multiplicative inverse [WAR 97]. Hamilton opened a door in mathematics to
hypercomplex algebras in general [STI 10, Chapter 20], [KAN 89], leading to the
octonions [CON 03, BAE 02] in less than a year, and the Clifford algebras about 30
years later [LOU 01, POR 95].

I.3. Quaternion Fourier transforms

Quaternion Fourier transforms, the subject of this book, are a generalization of

the classical Fourier transform to process signals or images with three- or
four-dimensional samples. Such signals arise very naturally in the physical world
from the three dimensions of physical space. Quite independently, for very different
(physiological) reasons connected with the trichromatic nature of human color vision
[MCI 98], color images have three components per pixel. The fourth dimension of
the quaternions plays a role in at least two ways: the frequency-domain
representation of a signal with three-dimensional samples requires four dimensions
(just as in the complex case, two dimensions are required in the frequency domain,
even if the original signal has one-dimensional samples). But more generally, the
four dimensions of the quaternions can be used to represent a most general set of
geometric operations in three dimensions using homogeneous coordinates, which are
explored in a later chapter (see section 2.3) and in [SAN 13a]. Of course,
generalizations to higher dimensions are possible, and there is a wide range of work
on Clifford Fourier transforms, which is outside the scope of this book (we refer the
readers to a recent volume for further details [HIT 13], and in particular the historical
introduction contained within [BRA 13]).
xviii Quaternion Fourier Transforms for Signal and Image Processing

I.4. Signal and image processing

Fourier transforms are a fundamental tool in signal and image processing. They
convert a signal or image from a representation based on sample or pixel amplitudes
into a representation based on the amplitudes and phases of sinusoids. The latter
representation is said to be in the frequency domain, and the original signal is said to
be in the time domain for a signal which is a time series, or in the image domain for
an image captured with a camera or scanner. Of course, signals may be encountered
that are not time series, for example, measurements of some physical quantity made
at (regular) intervals in space; in this book, we will use the terminology of time series
for simplicity, since the processing of other signals is mathematically no different.

The Fourier transformation is invertible, which means that the original signal or
image may be recovered from the frequency domain representation. More
interestingly, the frequency domain representation may be modified before inversion
of the transform, so that the recovered signal or image is a modified version of the
original, for example, with some frequencies or bands of frequencies suppressed,
attenuated or amplified. In some applications, inversion of the transform is not
needed: the processing performed in the frequency domain directly yields
information that can be immediately utilized. An example is computer vision, where
a decision based on analysis of an image may result in an action without any need to
construct an image from the processed frequency domain representation. At a more
detailed level, another example includes correlation, where the signal or image is
processed in the frequency domain to yield information about the location of a
known object within an image (the same applies in signal processing to find a known
signal occurring within a longer, noisy signal).

The classical Fourier transform is inherently based on complex numbers. This is

obvious from the fact that the frequency domain representation must represent both
the amplitude and the phase of each frequency present in the signal or image. The
symmetry of the transform means that the signal may be complex without any
modification of the transform. (There are some specialized variants of the Fourier
transform that handle only real signals, for example the Hartley transform
[BRA 86]). Given a signal with three components (representing, for example,
acceleration in three mutually perpendicular directions), how can a frequency domain
representation be calculated? The question is very similar if one considers a color
image: is it possible to construct a holistic frequency-domain representation of the
entire image? Obviously one can compute separate classical (i.e. complex) Fourier
transforms of the three components in both of these cases, but one then has three
separate frequency-domain representations, each representing one aspect of the
original image (the frequency content of one of the color or luminance/chrominance
channels). Processing of separate representations is sometimes known as marginal
processing, for reasons connected with techniques in the hand computation of
marginal distributions in statistics [TRU 53, section 1.22]. It is axiomatic in this book
 Introduction xix

that marginal processing is not the best way to handle signals and images with more
than two components per sample, but we will attempt to justify this belief throughout
the book, by showing how holistic approaches with quaternions yield better results.

There is another reason for using a quaternion Fourier transform in some

applications, and it provided the motivation for the earliest published work on
quaternion Fourier transforms (in the field of nuclear magnetic resonance (NMR)).
When a two-dimensional signal is captured (that is samples are measured over a
two-dimensional grid, like an image), it is sometimes necessary to regard the two
dimensions of the sampling grid as independent time-like axes. Processing such a
signal with a classical two-dimensional complex transform mixes the two
dimensions, whereas a suitably formulated quaternion transform does not. This is
because it is possible to associate each of the time-like dimensions with a different
dimension of the four-dimensional quaternion space, thus keeping the
frequency-domain representations of the first and second time-like axes apart. There
were two independent (as far as we are aware) early formulations of quaternion
Fourier transforms, by Ernst [ERN 87, section 6.4.2] and Delsuc [DEL 88,
equation 20], which are almost equivalent (they differ in the relative placement of the
exponentials and the signal, and in the signs, inside the exponentials):

∞ ∞
F (ω1 , ω2 ) = f (t1 , t2 )eiω1 t1 ejω2 t2 dt1 dt2 [I.1]
−∞ −∞

Note that the two-dimensional signal f (t1 , t2 ), is scalar-valued (i.e. not

quaternion-valued as in most of the cases discussed in this book). The two time-like
axes t1 and t2 are treated using separate quaternion roots of −1 (i and j), and
therefore they are not mixed. This may also be regarded as being due to the
orthogonality of the imaginary parts of the two exponentials.

Similar considerations motivated Thomas Bülow [BÜL 99, BÜL 01] when he
processed grayscale images using a quaternion Fourier transform. By using a
transform with samples of dimension greater than 2, Bülow was able to study
symmetries present in certain images in a way that is not possible with the
two-dimensional complex Fourier transform.

I.5. Other hypercomplex algebras

Of course, there are alternatives to quaternions for the construction and

computation of Fourier transforms for the type of applications suggested so far in this
introductory chapter, and it is worth briefly reviewing them here in order to provide a
full picture. As already noted, the quaternion algebra is not the only hypercomplex
algebra. In fact, the quaternion algebra is a specific example of a more general class
of hypercomplex algebras discovered by William Kingdon Clifford in 1876 (about 30
xx Quaternion Fourier Transforms for Signal and Image Processing

years after Hamilton’s discovery of quaternions) and known as Clifford algebras. The
Clifford algebras include the complex numbers and quaternions, but not the
octonions, curiously. However, as already mentioned, the quaternions share with the
real numbers and the complex numbers a very specific property that sets them aside
from all other Clifford algebras: every non-zero quaternion, q = 0, has a
multiplicative inverse such that q q −1 = q −1 q = 1. Furthermore, the quaternion
algebra is normed. This means that it is possible to define a norm (representing the
squared length of the quaternion in four dimensions) such that the norm of a product
of two quaternions equals the product of the norms of the two quaternions taken
separately: pq = p q . This is discussed in section 1.2 (see [1.14]).
Hypercomplex algebras in general have other troublesome properties. Many contain
values that are idempotent or nilpotent. An idempotent value q squares to give itself:
q 2 = q; and a nilpotent value squares to give zero: q 2 = 0. Such values obviously
have the potential to cause problems in numerical algorithms [ALF 07], and the
choice of the quaternion algebra avoids these problems entirely, because there are no
nilpotent or idempotent quaternions other than 0 and 1, respectively.

The one property of the quaternions that cannot be avoided is that multiplication
of quaternions is not commutative. This means that pq gives a different result in
general from qp for two arbitrary quaternions p and q. The reason for this can be
stated quite simply – the vector (or cross) product in three dimensions is not
commutative, and the product of two quaternions contains a vector product. It is
important to understand that this is an inherent property of three-dimensional
geometry and is not specific to the quaternions. This is again discussed in Chapter 1.
Non-commutative multiplication can be avoided by choosing a different
hypercomplex algebra, but since any hypercomplex algebra which is commutative
contains divisors of zero (a consequence of the Frobenius [DIC 14, section 11] ), any
attempt to avoid non-commutative multiplication will inevitably lead to other
problems, which may well be more troublesome than non-commutativity.
Non-commutative multiplication also occurs in linear algebra, of course, where the
product of matrices is dependent on ordering; so it should not cause undue concern to
anyone contemplating using quaternions.

I.6. Practical application

The ideas and concepts in this book are realisable in practice in several different
ways, particularly using software.

I.6.1. Software libraries

The library [SAN 13b] permits experimentation with transforms and other
algorithms operating on three- and four-dimensional data in MATLAB®. Since
 Introduction xxi

MATLAB® can generate C and C++ code (with some restrictions on supported
language features), quaternion code can, in principle, be used to generate code for
stand-alone applications, subject to licensing1.

Alternatively, code can be custom-written, using a quaternion library such as the

Boost library for C++, which contains some quaternion functions in the Math toolkit
[BRI 13]. This is discussed again in section 3.3, particularly with respect to the use of
decompositions into complex transforms to avoid the need to code elaborate
algorithms directly in quaternion code.

Both the QFTM and Boost libraries adopt the approach of directly coding
quaternion operations, that is they represent quaternions as quadruplets of real (or
complex) values, and provide elementary functions to add and multiply quaternions,
implementing the famous rules for ijk given in section 1.1, directly in code.

I.6.2. Matrix representations

An alternative to the use of quaternion libraries is possible, using matrix

representations, which we discuss here, and will return to with a very practical
application (to verification) in section 3.3.3.1.

Hypercomplex algebras (with the exception of the octonions [CON 03, BAE 02],
which are not associative) have matrix representations. What this means is that for a
given algebra, there exists a matrix algebra with real or complex elements that is
equivalent to the given hypercomplex algebra, in the sense that multiplication (and
addition of course) of the matrix representations is equivalent to multiplication in the
hypercomplex algebra. There are also other equivalences, for example the norm of a
hypercomplex value may be equivalent to the determinant of the matrix
representation. The matrix algebra using the given representation is said to be
isomorphic to the hypercomplex algebra. Matrix representations for the quaternions
are discussed in section 1.4.3, but we discuss the ramifications here.

Given the existence of a matrix representation of quaternions, it is theoretically

possible to substitute matrix representations for quaternions, both in algebraic
manipulation and in computer coding (the same is true for other hypercomplex
algebras except the octonions). Doing so can be a useful technique in theoretical
development because it can reduce a hypercomplex problem to a problem involving
real (or complex) matrices, and thus provide a deep insight into the relation between
the hypercomplex case and the well understood real and complex cases. However,
there are disadvantages of using a matrix representation compared to a direct
quaternion approach, in practice:

1 QTFM is licensed under the GNU General Public License.

xxii Quaternion Fourier Transforms for Signal and Image Processing

1) Computation with matrices is numerically inefficient, and it requires four times

as much memory as a direct quaternion representation storing only four values.
A quaternion product requires 16 multiplications and 12 additions, whereas the
equivalent 4 × 4 matrix product requires 64 multiplications and 48 additions, an
increase by a factor of 4. This disadvantage effectively rules out the use of matrix
representations for implementation, even for computer simulations (as run-time is four
times less using a quaternion library that directly calculates quaternion products than
using a general matrix package and a matrix representation for each quaternion).
2) The result of a sequence of arithmetic operations in matrix form may not be an
accurate representation of a quaternion matrix. This disadvantage again rules out the
use of matrix representations for implementation (a quaternion library will yield more
accurate results).
3) The matrix representation provides little geometric insight. As will be shown
in Chapter 2, the quaternion algebra provides a remarkably intuitive link with the
geometry of three or four dimensions. It is possible to manipulate quaternion symbols
algebraically in order to derive expressions for geometric operations. This is the
central idea in geometric algebras [SOM 01]. Geometric algebras, as might be
expected from the preceding text, are not division algebras; so there is a price to be
paid for their additional geometric utility in the form of divisors of zero, which make
them less attractive for applications in digital signal and image processing.

The matrix representation is certainly useful, and it is helpful in any quaternion

library to have the ability to convert between a direct quaternion representation and
the matrix representation. In the QTFM library [SAN 13b], for example, functions
called adjoint2 and unadjoint are provided to perform the conversion, even for
matrices of quaternions (the adjoint in the latter case is a block matrix with each
block representing one quaternion).

I.7. Overview of the remaining chapters

The rest of the book is divided into four chapters. Chapter 1 covers the quaternion
algebra, and provides the mathematical definitions and concepts necessary for the
later chapters. Chapter 2 presents the geometric applications of quaternions, and
provides the ideas necessary to understand how quaternions can be used to represent
both three- and four-dimensional values and geometric operations applied to them.
Chapter 3 gives a detailed and comprehensive account of quaternion Fourier
transforms, including their definitions, operator formulas and how they may be
computed. Chapter 4 shows how quaternion Fourier transforms can be applied in
signal and image processing.

2 The “adjoint” terminology was taken from Zhang’s 1997 paper [ZHA 97].
 1

Quaternion Algebra

This chapter introduces the quaternion algebra H and presents some properties that will be useful
in later chapters.

1.1. Definitions

Quaternions are one of the four existing normed division algebras over the real
numbers. Classically denoted by H in honor of Sir W.R. Hamilton who discovered
them in 1843, they form a non-commutative algebra. A quaternion q ∈ H is a four-
dimensional (4D) hypercomplex number and has a Cartesian form given by:

q = a + ib + jc + kd [1.1]

where a, b, c, d ∈ R are called its components. The three imaginary units i, j, k are
square roots of −1 and are related through the famous1 relations:

i2 = j 2 = k2 = ijk = −1
ij = −ji = k
[1.2]
ki = −ik = j
jk = −kj = i

A quaternion q ∈ H can be decomposed into a scalar part S(q) and a vector part
V(q):

q = S(q) + V(q) [1.3]

1 These relations defining the three imaginary units of an element of H were carved by Sir W. R.
Hamilton on a stone of the Broome bridge in Dublin on 16 October 1843.
Random documents with unrelated
content Scribd suggests to you:
its author. Thus he says—and the philosophic tone of the words is of
interest in view of the neglect and poverty into which he is said to
have fallen in his last years: “No one is content with his possessions,
and in the end eight feet of earth suffice us and there ends and is
consumed the vanity of our high thoughts,” and “Virtuous men who
love God and are of clean heart and uncovetous are never forsaken
of the grace of the Holy Spirit.”
He dwells more than once on the iniquity of oblivion wrought by
time: “Difference of ages and length of time hide the knowledge of
things and render them forgotten.” His descriptions are clearly those
of an eyewitness, as that of “a little river which flows from the top of
the mountains to the sea through reeds and mint and rushes and
wild-olives.” He praises Prince Henry the Navigator and King João II,
whose deeds are worthy to be told “by the ancient fathers of
eloquence and learning,” and it was in gratitude to them, a gratitude
which posterity shares, that he wrote: “Experience causes us to live
free of the false abuses and fables that some of the ancient
cosmographers recorded.”
Although the great events of India under the rule of Albuquerque
may have obscured the deeds of Pacheco, he was evidently not
forgotten, for in January 1509 he was sent with several ships against
the French pirate Mondragon and defeated and captured him off
Cape Finisterre, and probably about the year 1520 he was appointed
Governor of the fort of São Jorge de Mina, a coveted post on the
west coast of Africa.
Tradition has it that he came home in irons, and he may have
been the victim of one of those accusations by subordinates which
were becoming so common in the Portuguese overseas
possessions. Pacheco had shown of old that he was one of those
whom he calls inimigos da cobiça, with thoughts set on higher things
than gold. But a new king was on the throne, who was but two years
old when Pacheco was winning immortal renown for the Portuguese
in India, and it seems to have been the general feeling that he was
unfairly treated. Camões speaks of his “harsh and unjust reward.”
 It appears that he continued to receive his pension, yet he is said
to have died, about the year 1530, in extreme penury. We may be
sure at least that his heart did not quail before poverty any more than
it had before the countless host of Calicut. The recollection of his
wiles and devices during those hundred days at Cochin must have
been a powerful antidote to neglect and old age. “The thought of
what he had done would prove music to him at midnight.”
A few, no doubt, of the heroic ninety on whose behalf Pacheco
wrote to the King, recalling their services, survived, and they might
discuss the apparent miracle of their famous victory, and, in
Pacheco’s words, “the multitude of things in the very wealthy
kingdoms of India,” glad at heart the while to be at home under the
more temperate sun of Portugal and to rind their “eight feet of earth”
in their own soil.

FOOTNOTES:
[10] “The nobles,” says Correa, “are called Nairs, and are men
devoted to war.” The peasants “are so accursed that if they go
along a road they must go shouting, lest Nairs should meet and
kill them, for they may not carry arms, whereas the Nairs are
always armed. And if as they go shouting a Nair answers they
scuttle away into the wilds far from the road.”
[11] The poet Luis de Camões, after his return from the East,
supported life on less than a third of that amount.
 AFFONSO DE ALBUQUERQUE.
From Gaspar Correa, Lendas da India, frontispiece to vol. ii. pt.
1.
 VI
AFFONSO DE ALBUQUERQUE
(1462?-1515)
Aquelle invencivel e espantoso capitão Affonso de
Albuquerque.—Heitor Pinto, Imagem da Vida Christam.
O sem segundo Affonso de Albuquerque, honra de todos
os advertidos e scientes capitães que teve o mundo.—João
Ribeiro, Fatalidade historica da Ilha de Ceilão.
Albuquerque terribil, Castro forte.—Camões, Os Lusiadas.
Had Affonso de Albuquerque died five or six years before he did
the world would never have realised that it had lost one of the
greatest men of all nations and ages. Born of an ancient family[12]
about the year 1460,[13] Albuquerque had in 1514 seen thirty-eight
years’ service. He won the regard of Prince João in the campaign
against Spain in which that prince saved his father from irretrievable
defeat, and he became his equerry when he had succeeded to the
throne as João II. He also served with distinction in Africa.
It was in 1503, when he was over forty, that he first went to India.
In April of that year he sailed with his cousin Francisco de
Albuquerque in command of six ships, the chief object of the
expedition being to establish the friendship existing between the
King of Cochin and the Portuguese and to build a fort at Cochin.
Albuquerque made no long stay in India, and in July of the following
year was back in Lisbon. But he remained long enough to see the
vast possibilities there of failure or success for Portugal, and when,
two years later, he again went out, although he sailed as the
subordinate of Tristão da Cunha, it was on the understanding that he
should soon obtain independent command and with the provisional
appointment as Governor of India in his pocket.
 Smooth co-operation with other officials was not Albuquerque’s
strong point, and he felt no doubt that if he was to serve his King and
country as he would wish he must be able to act freely. It is
significant of his commanding personality that during his two years’
presence at Court he succeeded in imposing his views. In his
absence later his enemies were often able to tie him hand and foot
even though he was Governor of India.
There were two opposed policies. Hitherto the Portuguese in India
had been confined to the sea, and many considered that this
situation should continue. In a sense they were right, since it was
obviously impossible in so vast an empire to conquer and hold large
tracts of land. But Albuquerque considered that this floating empire
should be nailed down at cardinal points by capturing important
towns and building strong forts, and it was with this purpose that he
went out to India.
In the summer of 1507 he separated, according to his instructions,
from Tristão da Cunha, and when the latter returned to Portugal with
the rest of the fleet Albuquerque with his six ships remained in India.
Of these ships he has left a vivid description: there were no
provisions, the lances and other arms were few and rotten, with
great scarcity of cables, sails, and rigging; the powder was all wet, of
bombardiers there were but few, of carpenters one or two, and a
hundred and fifty men were dying of disease.
Even so he set to work to strike terror into the Moors and hammer
the Portuguese Empire into shape. Coasting down Arabia he sacked
various cities, spreading desolation with fire and sword and
mercilessly mutilating the Moors who surrendered. The poet Antonio
Ferreira called Albuquerque “clement.” It is not a clemency that we
would wish to encounter in ordinary life, and even among his
contemporaries some condemned his cruelty. Bishop Osorio, for
instance, considered it as unworthy of so great a man: illius rebus
gestis indignum.
But although Albuquerque could be harsh and grim enough (his
suggestion to King Manoel that Spanish and Portuguese Jews in
India should be extinguished one by one is most sinister), and was
quick to anger and a stern disciplinarian, he had no delight in cruelty
for cruelty’s sake. He wished to reduce the Moors throughout India to
subjection, and considered that such acts would best spread the
terror of his name and conceal the difficulties of his position. He
would have been the first to admit that his policy in this respect was
a sign of weakness.
Albuquerque’s first great achievement was the bombardment and
capture of the important city of Ormuz, at the entrance of the Persian
Gulf, and in October he set about building a fortress. Milton in the
following century wrote of “the wealth of Ormuz.” To Albuquerque it
was but the first stone in the vast edifice of his projects, but to his
captains it was already more than enough. They wished to be
making prizes on the high seas, not to be bottled up in Ormuz
building a fort as if they were masons. Albuquerque, to whom in their
complaints they were very much like gnats in a thunderstorm, went
on with his work, tore up their first petition and placed a second
under a jamb of one of the fort’s doorways as it was being built. This
was too much for the vanity of his captains and several of them
sailed away to India.
The result of this desertion was that Albuquerque was obliged
temporarily to abandon Ormuz. Small wonder that he wrote of their
conduct with extreme bitterness. “Without shame or fear of the King
or your Lordship,” he says in his letter to the Viceroy, Dom Francisco
de Almeida, “they deserted me in time of war, and during actual
hostilities with this city they left me and fled.... Portuguese gentlemen
have been guilty of no such vileness these three hundred years, nor
have I read of any such in the ancient chronicles.”
Even if they had all the right in the world on their side, these men
had deserted in the presence of the enemy; and had they been shot
by order of the Viceroy there and then, Portuguese rule would have
been greatly helped and strengthened and not only many troubles
but many lives spared in the future.
But no such salutary discipline prevailed in India; the instructions
given to the captains were partly independent, and the Viceroy
received them courteously and bade them draw up a document of
their complaints. When Albuquerque arrived in India his enemies
took care to foster differences between him and the Viceroy, who
was opposed to Albuquerque’s policy and methods, and, after being
treated with great discourtesy, Albuquerque was placed under arrest.
One of the accusations of his captains was that he wished to make
himself King of Ormuz.
They little knew their man. To expect Albuquerque, whose dreams
of conquest were as wide and magnificent as those of Alexander, to
vegetate as King of Ormuz was a mistake as colossal as to believe
that Napoleon could be content to rule Elba. There can be no doubt
that Albuquerque was unjustly treated by men incapable of
understanding him, all the more so in that Almeida’s term of office
was up and by right it was Albuquerque and not he who should have
been governing India. Albuquerque for his part disdained to be
conciliatory.
Fortunately for Albuquerque and for India his imprisonment only
lasted a few weeks. The arrival of the Marshal, Fernando Coutinho,
from Portugal put a new face on the situation; he released
Albuquerque and installed him as Governor of India. Almeida set out
for, but never reached, Portugal.
The year 1509 was almost out, and it is 1510 which marks the
beginning of Albuquerque’s victories. With the Marshal he attacked
Calicut, but the Marshal’s impetuous rashness (he was so nettled by
a first success of the impetuous but wise Albuquerque that he said
he would take Calicut with no other arm than a stick in his hand)
involved the expedition in disaster, and, although they sacked
Calicut, the Marshal and many of the Portuguese lost their lives in a
disorderly retreat to the ships, Albuquerque himself receiving a
wound which permanently disabled his left arm.
The rest of the year was occupied with Goa.[14] He obtained
possession of this city after a mere show of resistance, but a large
and ever-growing army of Turks forced him to abandon it after being
reduced to great straits and danger. Albuquerque had had fresh
trouble with his captains, but on the arrival of a few ships from
Portugal he returned to Goa in the autumn and stormed it. Most of
the Moors were put to the sword in a massacre which lasted four
days. Some Moorish women of almost white complexion he married
to Portuguese soldiers. This was a deliberate policy, approved by the
King of Portugal, in order to provide a peaceful settled population.
The possession of Goa changed the whole position of the
Portuguese in India. Remote kings who had hitherto looked on the
new-comers as passing freebooters now sent ambassadors offering
friendship and treaties.
Barely six months after taking Goa, Albuquerque stormed and
sacked Malaca, in Malay, a city which now belongs to the British
Empire and has about 100,000 inhabitants, and which then, in
Albuquerque’s own words, was “muito grande cousa.”[15] Of all the
great spoils the Governor characteristically reserved for himself only
two great bronze lions which he intended to have placed on his
tomb. But his ship, laden with the costliest plunder, much of which
was intended for King Manoel, met with a violent storm and
foundered. Albuquerque, dressed in a brown coat and anything that
came to hand, escaped on a raft.
In 1513 he carried out his long-cherished project of an attack on
Aden, whence, he said, “vermilion, currants, almonds, opium,
horses, dates, gold” went to India. The Portuguese assaulted but
failed to take the town—in their eagerness the ladders broke again
and again under their weight—and it was not safe to blockade it for
fear of adverse winds, lack of water, and the large and speedy
assistance the enemy might expect. Swift cameleers carried the
news of the attack in fifteen days to Cairo, and, generally, the
presence of a large Portuguese fleet in the Red Sea made a far-
reaching impression.
Albuquerque set out to attack Aden again in 1515, but was
occupied for some time at Ormuz, and fell ill there. He started to
return to India, and on the way received tidings from a passing boat
that his successor to the Governorship of India had been appointed,
and many important posts given to his personal enemies.
This was his death-blow. Only a year before he had written to the
King of his determination to continue in India for the rest of his life, at
whatever sacrifice to himself, for the sake of maintaining and
strengthening the empire he had won. Now heartbroken he
exclaimed, “Out of favour with men for the sake of the King, and out
of favour with the King for the sake of men. It is good to make an
end.” He dictated a last brief letter to the King “in the throes of
death,” recommending his son, and died as the ship came in sight of
Goa, straining his eyes to see the tower of the church he had
founded (December 1515).
Next day his body, dressed in the habit of Santiago, was carried
ashore and buried amid universal grief. The natives perhaps
mourned him sincerely, since he had worked for their prosperity and
his attitude towards them, as distinguished from the Moors, had
always been kindly. The gods, they said, had summoned him to war
in heaven. His enemies continued to fear him even dead, so that
King João III declared that India would be safe so long as
Albuquerque’s body remained there, and it was only in 1566 that his
bones were brought to Portugal.
A contemporary Portuguese historian, Barros, thus describes
Albuquerque: “He was a man of medium height, of a cheerful,
pleasant countenance, but when angry he had a melancholy look; he
wore his beard very long during the time of his command in India,
and as it was white it made him very venerable. He was a man of
many witty sayings and in some slight annoyances [menencorias
leves! Had not Barros read Albuquerque’s letters?] during his
command he said many things the wit of which delighted those
whom they did not immediately affect. He spoke and wrote very well
with the help of a certain knowledge of Latin [the superior Barros!].
He was cunning and sagacious in business, and knew how to mould
things to his purpose, and had a great store of anecdotes suited to
different times and persons. He was very rough and violent when
displeased and he tired men greatly by his orders, being of a very
urgent disposition. He was very charitable and devout, ever ready to
bury the dead. In action he was somewhat impetuous and harsh. He
made himself greatly feared by the Moors and always succeeded in
getting the better of them.”
 Another historian, Correa, who had served Albuquerque three
years as private secretary in India, knew him better and appreciated
his greatness. It is Correa who gives us an imposing glimpse of the
Governor of India two years before his death, i.e. at the time when
Albuquerque described himself as “a weak old man.” He was
dressed “in doublet and flowing open robe, as was then the fashion,
all of black damask streaked with black velvet, on his head a net of
black and gold thread, and above this a large cap of black velvet; in
his belt a dagger of gold and precious stones worth fifteen thousand
crusados, round his neck a thick chain; and his long white beard,
knotted at the end, gave him a very venerable presence.”
Albuquerque was sincerely devout, even to the verge of mysticism
or superstition. He believed that St. James went before the
Portuguese on a white horse guiding them to victory, and when in
the Red Sea that a fiery cross in the sky was specially sent to
beckon him on to further conquests.
There is a massive strength in all that he said and did.[16] After he
had subdued Ormuz its king hesitated whether he should pay his
customary tribute to Persia and sent to consult Albuquerque.
Albuquerque made a little collection of firearms and cannon-balls
and answered, “In this coin is the King of Portugal wont to pay
tribute.”
But the whole man is in his letters, aptly described as being
“written with a sword.” Perhaps it is only in the letters of Napoleon
that one finds the same mingling of great plans and conceptions with
a mastery of the smallest details and concern for things which a
lesser man would scorn to notice.
This Governor, the fear of whose name extended far into China, to
whom the Kings of Narsinga and Persia, Siam, Cambaya, Turkey,
and Cairo sent gifts, the conqueror of Ormuz and Cananor, Goa and
Malaca, who dispatched his agents even to the remote Moluccas,
and who was determined to destroy Mecca (five hundred Portuguese
were to ride swiftly inland from the coast, take it by surprise and burn
it to ashes) and thought of altering the course of the Nile, did not
disdain to occupy himself with the alphabets for teaching children to
read, the missals and pontificals for churches, pearl-fisheries, the
horse trade, the colour of the Red Sea, how to pack quicksilver, and
a hundred other matters of great diversity, while on the question of
arms and merchandise to be sent from Portugal to India[17] no
modern official report could exceed his letters in accuracy and
minuteness.
For instance, he declares that lances are sent out unsharpened,
as they come from the Biscay factories, to the care of a barbeiro
inchado in India, and in 1513 says that he now has workmen in Goa
who can turn out better guns than those of Germany. Unfortunately
in Portugal India was regarded merely as a mine to be exploited, not
as a field that required farming in order to continue productive.
Albuquerque, when, as he says, over his neck in work, had to
answer great bundles of letters from the King, often filled with
carping criticisms of his actions or containing contradictory projects.
He complains that there is a new policy for each year, almost in the
words of Dante in the Purgatorio:
fai tanto sottili
Provvedimenti ch’ a mezzo Novembre
Non giunge quel que tu d’Ottobre fili.
It must be confessed that Albuquerque in these letters, filled with
the eloquence of the Old Testament, gave as good as he got: the pity
is that the King probably only saw them in the official summaries.
“Sir, the soldiers in India require to be paid their salaries,” he says on
one occasion, or “Your Highness is not well informed,” and he warns
him that should matters continue as in the past the empire will come
crumbling about the King’s ears.
Again he writes that he is not amazed that the accusations should
be made, but amazed that the King should believe them. The names
of his accusers were withheld, as later in trials before the Inquisition,
but he knew whence the trouble came and does not mince his words
in telling the King of the corruption, greed, carelessness, and
incompetence of the officials in India appointed by the King. “And if I
were not afraid of Your Highness I would send you a dozen of these
mischief-makers in a cage.”
 In five days he writes nineteen letters to the King, some of them of
considerable length, this task occupying him till dawn, after a long
day’s work. On a single day he wrote the King eight letters, one of
which contains a splendid general account of the state of India,
another is a little masterpiece describing the misdeeds of one of his
captains.
No doubt his critics believed him to be harsh and insensible. That
this was far from being the case is shown by the fact that on
receiving, amid a shower of blame and criticisms, a sympathetic
letter from his old friend, the historian Duarte Galvão, he shed tears,
and also by the deep feeling he displayed when a whole batch of
letters came from the King full of dispraise. “Your Highness blames
me, blames me, blames me,” he wrote, and again, “My spirits fell to
the ground and my hair turned twice as white as it was before.”
When, therefore, a few months after he had written of his intention
to return from Ormuz to India in order to see the King’s letters and
know if he had sent ships and men for the expedition against Aden,
he heard that his successor was appointed, it is scarcely an
exaggeration to say that the news killed him.
Albuquerque’s crime was to have thought of India and Portugal
first, before personal interests and ambitions. “They call me a harsh
man,” he said; and “these officials of yours do not love me.” But if he
could vigorously show his dislike of the false and slovenly, he always
liberally rewarded good service, was loyal, generous, and unselfish,
and showed a most delightful pleasure in any thorough work or
workman. “The best thing I ever saw,” he says of a map; and of a
good carpenter, “he is a marvellous man.”[18]
No sooner was Albuquerque dead than his greatness was felt, and
posterity has never sought to deny it. If we consider the conditions
under which his great work was accomplished in six years—his ships
often so rotten that they sank of sheer old age, his men few and ill-
armed (before he received reinforcements in October 1512 he says
that the whole number of Europeans under his command in India
were but 1,200, of whom barely 300 were properly armed), the fact
that all his projects were liable to be upset by orders dictated in
ignorance at home, and that as soon as his back was turned (for
instance, when he went to attack Aden) all the officials in India
treated him as dead and his instructions as a dead letter—we will not
deny that posterity has done well to honour and admire this man in
his lonely magnificence. Fannomi onore e di ciò fanno bene.
No doubt he had great faults, since everything in him was great.
He adopted oriental methods in dealing with the kings of the East.
He murdered in cold blood the powerful minister of the young King of
Cochin, and in one of his letters to King Manoel he remarks calmly,
“In all my letters I bade him kill the Samuri of Calicut with poison.”
But he understood the East and was the only man who could have
established the Portuguese Empire firmly. That he was not given a
free hand and every assistance from the first was the doom of that
empire, and Portugal never saw his like again.

FOOTNOTES:
[12] Albuquerque’s father, Gonçalo de Albuquerque, was in
favour at Court. His grandfather João Gonçalvez had been
secretary to King João I and King Duarte, but was hanged for
murdering his wife in 1437.
[13] In 1461 or 1462. In one of his letters (April 1, 1512) he
says that he is fifty. Correa, who calls him old in 1509, says that
he was over seventy at the time of his death. Despite the very
definite assertion in his letter, perhaps the last word has not been
said as to his age. Misprints in these matters are common. Couto,
for instance, says that Albuquerque’s nephew Naronha is nearly
seventy in 1538 and eighty in 1540. All the historians call
Albuquerque old, yet the captain of a fortress was considered too
young for the post because he was under forty (Correa III, 687).
On the other hand not Borrow merely but Couto (VI. 2. ix) calls
Castro old, although he did not live to be fifty. Perhaps in
Albuquerque’s letter we should read LX instead of L (for indeed
why should he speak so fatherly to King Manoel (1469-1521) if he
was not considerably older than the King?), and sesenta for
setenta in Correa.
 [14] Goa is thus described by an early traveller: “La città di Goa
è la più fresca delle Indie e la più abbondante di tutte le cose
da.... È detta città molto grande, con buone case e grandi e belle
strade e piazze, murata d’intorno con le sue torri e fatta in una
buona fortezza. Fuori di detta città vi erano molti horti e giardini
copiosi e pieni d’infiniti arbori fruttiferi, con molti stagni di acque;
eranvi molte moschee e case d’ orationi di gentili. Il paese
d’intorno è molto fertile e ben lavorato.”
[15] The same traveller says: “Questa città di Malaca è la più
ricca scala di più ricchi mercatanti e di maggior navigatione e
traffico che si possa trovare nel mondo.”
[16] The story, maliciously recorded by Barros, that
Albuquerque sent ruby and diamond rings to the historian Ruy de
Pina to jog his memory in relating the events of India, may or may
not be true. In a way it is characteristic, for Albuquerque, if he
wished for Pina’s praise, which one may be inclined to doubt, was
not a man to beat about the bush. Perhaps after all it was more
honest to plump down the rubies than to indulge in elogio mutuo.
[17] In one letter he bids the King plant all the marsh-lands of
Portugal with poppies, since opium is the most welcome
merchandise in India.
[18] Estimava muito os homens cavalleiros, says Correa, who
knew him personally and insists more than once that he was very
accessible. To cope with what Albuquerque himself calls the
“mountains of petitions” that beset him he employed six or seven
secretaries, but he dealt with them unconventionally, signing them
or tearing them up in the street as they were given him, thereby
expediting his business but offending the vanity of the petitioners.
JOÃO DE CASTRO.
 VII
DOM JOÃO DE CASTRO
(1500-1548)
Era tambem de sua pessoa tam esforçado como em letras
insigne.—Pedro de Mariz, Dialogos de Varia Historia.
In that low shady quinta, embowered amongst those tall
alcornoques, once dwelt John de Castro, the strange old
Viceroy of India.—George Borrow, The Bible in Spain.
Castro was still a schoolboy when Albuquerque died. Born in
1500, the son of D. Alvaro de Castro, in high office under Kings João
II and Manoel, and a daughter of the Count of Abrantes, he studied
with the famous mathematician, Pedro Nunez, and had a scientific
as well as a classical education. There is every reason to believe
that he was a promising and fervent scholar, but the victories of Dom
Duarte de Meneses in North Africa appealed to him even more than
did the figures of Euclid, and in 1518 he “took the key of the fields”
and fled to Tangier. There he served with the greatest distinction for
nine years, and stood high in favour with the Governor, Meneses,
who knighted him and on his return to Portugal in 1527 furnished him
with a glowing recommendation to the King.
Of the next few years of his life comparatively little is known. He
received a comenda from the King, was employed on various
service, and married D. Leonor de Coutinho, of noble family but
poor. Probably he was able to devote considerable time to quiet
study. In 1535 he commanded one of the twenty-five Portuguese
ships in the Emperor Charles V’s victorious expedition against Tunis.
It was on this occasion that Castro’s lifelong friend, the gallant poet
Prince Luis, followed his example of 1518 and ran away to join the
expedition against the wishes of his brother King João III.[19]
 In the autumn Castro was back in his favourite Cintra. There he
himself planted a quinta, to which his thoughts, later in India,
constantly turned. Those who go along the delightful shady road of
orchard and running streams, rock and woodland from Cintra to
Monserrate and Collares come in a few minutes to an archway and
green door on the right. It is here, in the quinta now known as Penha
Verde, overlooking the fertile plain of Collares to the sea, that Castro,
like Pitt planting by moonlight or Garibaldi in his island, indulged his
love of husbandry.
“Here,” says one of his early biographers, “he entertained himself
with a new and strange kind of agriculture, for he cut down fruit-
bearing trees and planted wild woods, perhaps to show that he was
so disinterested that not even from the earth would he expect
reward. Yet it is no wonder if one who disdained the rubies and
diamonds of the East should think little of the products of Cintra’s
rocks.” It was to the matos of the Serra de Cintra that he longed to
return in 1546. But he certainly did not despise the fruits of the soil,
and probably occupied himself with grafting experiments.
In the spring of 1538, as perhaps previously in the spring of 1537,
he sailed to India as captain of a ship. The fleet arrived at Goa in
September 1538 and went on to the relief of Diu. In March of the
following year he returned to Goa, and two years later accompanied
the new Governor, Dom Estevão da Gama, to the Red Sea.
On all these occasions Castro kept a log or roteiro, from Lisbon to
Goa, from Goa to Diu, and from Goa to the Red Sea. They display a
strong scientific interest, a spirit thoroughly modern—nothing,
however small it might be, was to him necessarily unimportant or
negligible—or perhaps ancient, since he complains that in his day
the scientific investigations of the ancients were no longer in vogue.
The logs are written with that vivid directness which mark his letters,
“written,” he said, “not for the ladies and gallants of the Court and
royal palaces, but for the mariners of Leça and Mattosinhos.”
His descriptions are precise and accurate, which does not prevent
them from being often picturesque. He notices many birds, including
one white and grey which, he says, the sailors call frades (monks). “I
pay great attention to eclipses of the moon,” he writes, as also to
longitudes and latitudes, fishes, seaweeds, currents, winds, the
colour of the Red Sea, and every detail that might concern the art of
navigation, to the delight of his friends Dr. Pedro Nunez and Prince
Luis, who had furnished him with special instruments and other
assistance for his voyage.
In the summer of 1542 he was back at Cintra, but in December of
that year he was appointed to the command of the coast fleet, the
main duties of which were to keep clear the coast of Portugal from
pirates, such as Mondragon, who perpetually hovered in wait for the
priceless spoils and cargoes of Portuguese ships homeward bound
from India. He seems to have gone to sea before the end of the year
and held this post for two years, with a brief interval in 1543 when he
commanded the Portuguese fleet sent to co-operate with the
Spanish against Barbarossa. They did not come to an engagement,
and Dom João, after visiting Ceuta, returned to Portugal.
He was at Cintra in the beginning of 1545 when the unwelcome
news reached him that he had been appointed Governor of India.
Most unwillingly he accepted this new post, the difficulties and
disquiet of which he had been able to gauge at first hand during his
former sojourn in Goa. His young sons were to accompany him.
A picturesque story of the Governor-elect cannot be better told
than in the words of the historian Couto, who served under him in
India: “Passing one day by the door of a tailor [in Lisbon] he noticed
a pair of very rich and fashionable velvet breeches, and pulling up
his horse asked to see them. After examining their curious
workmanship he asked whose they were. The tailor, not knowing
whom he was addressing, answered that they were for a son of the
Governor who was going to India. Dom João de Castro thereupon in
a rage took up a pair of scissors and cut them into shreds. “Bid that
young man buy arms,” he said to the tailor, and so passed on.”
At the end of March the fleet sailed. The number of men actually
enlisted was eight hundred, but many more who had been rejected
for some defect or were escaping from justice succeeded in
embarking as stowaways. In the Governor’s ship alone there were
nearly two hundred of them, and they required to be fed during a
voyage of many weeks. The Governor was advised to cast them
adrift in the provision ship or to maroon them in the Cape Verde
Islands, but humanely and persistently refused.
He had not been long at Goa when, in April 1546, news was
brought that a formidable attack was being prepared against Diu, the
fort commanded by the heroic Dom João de Mascarenhas. Castro
sent his son Dom Alvaro with a strong fleet to its relief. The fleet was
delayed by violent storms, and when it finally reached Diu there was
little of the fortress left. The walls and bulwarks were levelled with
the ground, most of the defenders dead, and those who remained
either wounded or ill. No one but Mascarenhas could have held on in
such conditions, and even so “six more days,” wrote Castro to the
King, “and relief would have come too late.”
Most of the nobles in Diu were dead, and among them Dom João
de Castro’s other son, Fernando, who had been blown up with many
others on a mined part of the wall on which they had rashly
remained, although warned by Mascarenhas of their danger. “He
should have obeyed Dom João,” wrote Castro stoically to the King,
and he added: “Of what Dom Fernando did till the time of his death I
will say nothing to your Highness, for it cannot be that men are so
wicked but that some among them will inform your Highness of the
services and great exertions that my sons undergo in your service.”
The King of Cambaya still boasted of victory, and Dom João de
Castro himself sailed north with a powerful fleet from Goa. After
striking terror into the enemy by ravaging the coast of Cambaya,
setting it all aflame and, in his own words, “sparing no living thing,”
he left these shores covered with dead and crossed to Diu.
The fortress was now again invested by an army of 60,000 Moors,
and in the battle with the besieging force the Governor was himself
more than once in the greatest danger before the enemy was routed.
Indeed, it was his personal exertions which largely decided the day,
and with pardonable pride he wrote to the King that it was “the
greatest victory ever seen in all the East.”